{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:FZMFAVZF5QPMYNFFQNT55CUI6T","short_pith_number":"pith:FZMFAVZF","schema_version":"1.0","canonical_sha256":"2e58505725ec1ecc34a58367de8a88f4c973a8eb3fbe1d73d5bc416cc196b314","source":{"kind":"arxiv","id":"2310.05344","version":1},"attestation_state":"computed","paper":{"title":"SteerLM: Attribute Conditioned SFT as an (User-Steerable) Alternative to RLHF","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Makesh Narsimhan Sreedhar, Oleksii Kuchaiev, Xianchao Wu, Yi Dong, Zhilin Wang","submitted_at":"2023-10-09T02:11:21Z","abstract_excerpt":"Model alignment with human preferences is an essential step in making Large Language Models (LLMs) helpful and consistent with human values. It typically consists of supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) stages. However, RLHF faces inherent limitations stemming from a complex training setup and its tendency to align the model with implicit values that end users cannot control at run-time. Moreover, reward models in RLHF stage commonly rely on single-dimensional feedback as opposed to explicit, multifaceted signals that indicate attributes such as he"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2310.05344","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-09T02:11:21Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a5c1921d0fde9e6d7a23ce1d8bb294f2546add8f273496cee3a8421ef5995d2c","abstract_canon_sha256":"40fc9fab9af34ae9b3319d85e5be1138e2dfa4c7cc3bded44c9cdf5114b7ac7b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:58:44.028540Z","signature_b64":"23s8Kr/wbSB8riAxIp3UkDx9lzM4pgLN0O/+xCy5T8WZ8Q9B9GeynQ4idq1cnSVJV2vub1KoUJL5Wzan6dGyCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2e58505725ec1ecc34a58367de8a88f4c973a8eb3fbe1d73d5bc416cc196b314","last_reissued_at":"2026-07-05T06:58:44.028042Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:58:44.028042Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SteerLM: Attribute Conditioned SFT as an (User-Steerable) Alternative to RLHF","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Makesh Narsimhan Sreedhar, Oleksii Kuchaiev, Xianchao Wu, Yi Dong, Zhilin Wang","submitted_at":"2023-10-09T02:11:21Z","abstract_excerpt":"Model alignment with human preferences is an essential step in making Large Language Models (LLMs) helpful and consistent with human values. It typically consists of supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) stages. However, RLHF faces inherent limitations stemming from a complex training setup and its tendency to align the model with implicit values that end users cannot control at run-time. Moreover, reward models in RLHF stage commonly rely on single-dimensional feedback as opposed to explicit, multifaceted signals that indicate attributes such as he"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.05344","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2310.05344/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2310.05344","created_at":"2026-07-05T06:58:44.028107+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.05344v1","created_at":"2026-07-05T06:58:44.028107+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.05344","created_at":"2026-07-05T06:58:44.028107+00:00"},{"alias_kind":"pith_short_12","alias_value":"FZMFAVZF5QPM","created_at":"2026-07-05T06:58:44.028107+00:00"},{"alias_kind":"pith_short_16","alias_value":"FZMFAVZF5QPMYNFF","created_at":"2026-07-05T06:58:44.028107+00:00"},{"alias_kind":"pith_short_8","alias_value":"FZMFAVZF","created_at":"2026-07-05T06:58:44.028107+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.27969","citing_title":"Boundary Suppression Asymmetry in Post-trained Assistants: Over-expansion as a Controllability Cost","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30873","citing_title":"Federated Variational Preference Alignment with Gumbel-Softmax Prior for Personalized User Preferences","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T","json":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T.json","graph_json":"https://pith.science/api/pith-number/FZMFAVZF5QPMYNFFQNT55CUI6T/graph.json","events_json":"https://pith.science/api/pith-number/FZMFAVZF5QPMYNFFQNT55CUI6T/events.json","paper":"https://pith.science/paper/FZMFAVZF"},"agent_actions":{"view_html":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T","download_json":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T.json","view_paper":"https://pith.science/paper/FZMFAVZF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.05344&json=true","fetch_graph":"https://pith.science/api/pith-number/FZMFAVZF5QPMYNFFQNT55CUI6T/graph.json","fetch_events":"https://pith.science/api/pith-number/FZMFAVZF5QPMYNFFQNT55CUI6T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T/action/storage_attestation","attest_author":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T/action/author_attestation","sign_citation":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T/action/citation_signature","submit_replication":"https://pith.science/pith/FZMFAVZF5QPMYNFFQNT55CUI6T/action/replication_record"}},"created_at":"2026-07-05T06:58:44.028107+00:00","updated_at":"2026-07-05T06:58:44.028107+00:00"}